You are on page 1of 3

Brief communication

Modeling anisotropic Reynolds-stress dissipation in particle-


or droplet-laden ows
Yvonne Reinhardt, Niklaus Meinen, Daniel W. Meyer

Institute of Fluid Dynamics, ETH Zurich, Sonneggstrasse 3, CH-8092 Zurich, Switzerland


a r t i c l e i n f o
Article history:
Received 9 March 2013
Received in revised form 3 May 2013
Accepted 4 May 2013
Available online 15 May 2013
Keywords:
Particles
Droplets
Turbulence modeling
Reynolds-stress modeling
RANS
a b s t r a c t
Particles or droplets dispersed in turbulent ows at sufciently high volume loadings lead to modica-
tions of turbulence characteristics. More specically, in a detailed experimental investigation by Poelma
and coworkers, where the particle phase moves with a non-zero mean velocity relative to the uid phase,
it was found that anisotropic Reynolds-stress dissipation is induced. Recently, we have proposed a model
that can account for this effect in RANS-based and PDF method simulations. In our previous work, how-
ever, no simulation results of the RANS model formulation were presented. In the present work, a new
compact tensorial RANS formulation is presented and the new formulation is validated against the exper-
imental data of Poelma and coworkers.
2013 Elsevier Ltd. All rights reserved.
1. Introduction
Particles or droplets dispersed in turbulent ows at sufciently
high volume loadings lead to modications of turbulence charac-
teristics (Balachandar and Eaton, 2010, Section 6). More speci-
cally, in the experimental investigation (Poelma et al., 2007),
where the particle phase moves with a non-zero mean velocity rel-
ative to the uid phase, it was found that anisotropic Reynolds-
stress dissipation is induced. This is similar to previous observa-
tions in the experiments (Geiss et al., 2004) and the direct numer-
ical simulation (DNS) study (Elghobashi and Truesdell, 1993).
Recently, Meyer (2012, Section 2.3) proposed a model that can
account for this effect in RANS-based and PDF method simulations.
In (Meyer, 2012), however, no simulation results of the RANS for-
mulation were presented. In the present work, a much more com-
pact tensorial RANS formulation compared to (Meyer, 2012) is
presented and the new formulation is validated against the exper-
imental data in (Poelma et al., 2007).
2. Formulation
The Reynolds-stress evolution equation for turbulent uid ow
with dispersed particles or droplets reads
@hu
0
i
u
0
j
i
@t
hu
k
i
@hu
0
i
u
0
j
i
@x
k

@hu
0
i
u
0
j
u
0
k
i
@x
k
P
ij
P
ij
m
@
2
hu
0
i
u
0
j
i
@x
k
@x
k
e
ij
hU
m
u
0
j
f
i
i hU
m
u
0
i
f
j
i 1
(Meyer, 2012, Eq. (2.13)). In Eq. (1), u
0
i
u
i
hu
i
i is the uctuating
part of the uid velocity component u
i
with mean hu
i
i subtracted,
hu
0
i
u
0
j
i is the ij-Reynolds-stress, t the time, x
k
a spatial coordinate,
P
ij
and P
ij
are the production and velocitypressure-gradient terms,
respectively, m is the kinematic viscosity, e
ij
the dissipation tensor,
and U
m
u
0
i
f
j

are extra-dissipation terms resulting from the particle
phase with the amount of dispersed particles being quantied by
the mass loading U
m
.
To model the anisotropy in e
ij
resulting from the particle phase
having a mean relative motion with respect to the uid phase, i.e.,
jhui hvij 0 with v being the particle phase velocity vector,
Meyer (2012) has introduced the dissipation tensor
21 j 0 0
0 j 1 0
0 0 j 1
2
6
4
3
7
5 I
0
B
@
1
C
A
2e
3
; 2
where I is the unit matrix, e the dissipation rate of the turbulent ki-
netic energy k and j can be interpreted as an anisotropy ratio with
j 1 C
a
U
m

jhui hvij

2k=3
p
s
g
d
: 3
The parameter C
a
is a model constant, g (m
3
/e)
1/4
is the Kolmogo-
rov length scale, and d the particle diameter of the monodisperse
0301-9322/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijmultiphaseow.2013.05.002

Corresponding author. Tel.: +41 44 633 9273; fax: +41 44 632 1147.
E-mail addresses: reinhardt@ifd.mavt.ethz.ch (Y. Reinhardt), meyerda@ethz.ch
(D.W. Meyer).
International Journal of Multiphase Flow 56 (2013) 13
Contents lists available at SciVerse ScienceDirect
International Journal of Multiphase Flow
j our nal homepage: www. el sevi er . com/ l ocat e/ i j mul ow
particle phase.
1
For hui = hvi, Eq. (3) leads to j = 1 whereas j > 1 for
hui hvi. To align the tensor of Eq. (2) with vector hui hvi, which
is causing the dissipation anisotropy, Meyer (2012, Eqs. (2.29) and
(2.31)) has applied a suitable rotation matrix.
In this work, we propose a mathematically more compact ten-
sorial model formulation with
e
ij
j 1d
ij
31 j
hu
i
i hv
i
ihu
j
i hv
j
i
jhui hvij
2
d
ij
" #
2e
3
; 4
where d
ij
is the Kronecker delta. Here, j acts as a blending factor be-
tween the standard isotropic model, i.e., e
ij

2
3
ed
ij
, and an aniso-
tropic tensor based on the relative velocity hui hvi. For example
with hui hvi being aligned with the x
1
-direction, e
ij
reduces to
Eq. (2). For j > 1, the rst two terms in the bracket of Eq. (4) lead
to a reduction of dissipation in the direction of hui hvi and an
enhancement of dissipation in the directions normal to hui hvi.
Tensors similar to Eq. (4) are for example used to model anisotropic
dispersion in subsurface ows (Bear, 1972, Eq. (10.4.15)).
3. Test case
To validate the new RANS model formulation for the dissipation
term, the experimental data of Poelma et al. (2007) is used as a ref-
erence. Poelma et al. (2007) have investigated the turbulent ow of
water loaded with solid monodisperse particles in the center of a
vertical channel, where wall effects are negligible. Their experi-
mental setup is sketched in Fig. 1. Both the uid and particle
phases move in x
1
-direction upward. A turbulence generating grid
with grid-spacing M = 7.5 mm is placed upstream of the test sec-
tion. At x
1
= 0, the velocity uctuations of the uid and particle
phases are equal and the uid phase Reynolds-stress tensor is iso-
tropic. The mean velocities of both phases are constant in space
and time. Due to the higher density of the particle phase, the aver-
age particle velocity hv
1
i is smaller compared to hu
1
i. The velocity
difference between the phases induces a growing Reynolds-stress
anisotropy in downstream direction with hu
0
1
u
0
1
i > hu
0
2
u
0
2
i hu
0
3
u
0
3
i
for x
1
> 0.
Based on different experiments, Poelma et al. (2007) have
investigated the inuence of particle characteristics and mass load-
ings on the uid phase Reynolds-stress anisotropy. The density ra-
tio between particle and uid phases was ranging from q
p
/q = 2.45
to 3.8, the Stokes number dened by the particle relaxation time
and the Kolmogorov timescale was between St = s
p
/s = 0.07 and
0.23 and mass loadings U
m
were between 0.18% and 0.67%. Details
about the particle and uid phase parameters are summarized in
(Meyer, 2012, Table 2).
4. Numerical implementation
For the small mass loadings considered by Poelma et al. (2007),
the extra-dissipation terms in the Reynolds-stress transport Eq. (1)
are negligible (Meyer, 2012, Section 3.2.3). To model the transport
term (rst term on the right hand side of Eq. (1)) the DalyHarlow
model (Pope, 2000, Eq. (11.147)) was used and the LRR-IP model
(Pope, 2000, Section 11.5.1) was applied for the velocitypres-
sure-gradient term. The latter includes Rottas return-to-isotropy
model (Pope, 2000, Eq. (11.24)). In the ow setup considered, the
Reynolds-stress isotropization of Rottas model is counteracted
by the dissipation model (Eq. (4)). To determine e appearing in
the different models, the standard empirical e-transport-equation
(Pope, 2000, Eq. (11.150)) is solved.
For the model parameters, the following standard values were
applied: in the e-equation, C
e
= 0.15 and C
e1
= 1.44, and in the mod-
eled Reynolds-stress transport equation, C
s
= 0.22 (DalyHarlow
model) and C
2
= 3/5 (IP term in LRR-IP model). In agreement with
the choice of C
x2
in (Meyer, 2012, p.268), C
e2
= 1 + C
x2
= 1.96
(Pope, 2000, Eq. (12.184)) was applied in the e-equation. This
choice deviates somewhat from the standard value of 1.92 but
leads to optimal agreement of the downstream decay of k (Meyer,
2012, Fig. 7). For the model constants in Rottas return-to-isotropy
model and Eq. (4), two value pairs, i.e., C
R
= 1.8 (standard value)
with C
a
= 14 and C
R
= 4.15 with C
a
= 28 (Meyer, 2012, p. 268) were
used. C
a
= 14 in the rst parameter pair, which was not used in
(Meyer, 2012), leads to optimal agreement with the experimental
Fig. 2. Validation of single-phase ow simulation results (lines) with experimental
data of Poelma et al. (2007, Table 2) (solid circles). The normalized inverse of the
turbulent kinetic energy (top plot) and the turbulence frequency (bottom plot) are
compared. The values reported by Poelma and coworkers for the turbulence
frequency or more precisely the dissipation rate needed to be corrected by a factor
of 2/3 (Poelma 2011, personal communication).
Table 1
Summary of RANS model coefcients with the two parameter pairs involving C
R
and
C
a
.
Parameter pair
One Two
C
e
C
e1
C
e2
C
s
C
2
C
R
C
a
C
R
C
a
0.15 1.44 1.96 0.22 3/5 1.8 14 4.15 28
1
A possible extension of j applicable for polydisperse particles is discussed in
(Meyer, 2012, Eq. (4.1)).
gravity
0
v
1

u
1
=
0.53m/s
x
2
x
mean flow
u water
v particles
m
u
v
1
x
u
1
u
1
=u
2
u
2

=v
1
v
1
=v
2
v
2

u
2

v
2

u
1
u
1
u
2
u
2

2

Reynolds
stresses
Fig. 1. Schematic of the ow setup applied in the experiments by Poelma et al.
(2007).
2 Y. Reinhardt et al. / International Journal of Multiphase Flow 56 (2013) 13
data (Poelma et al., 2007). As is explained next, the second param-
eter pair is consistent with the PDF method simulations reported in
(Meyer, 2012, Section 3.2.3). Based on the simplied Langevin
model (SLM) parameter C
0
= 2.1, which is typically applied in PDF
methods (Pope, 2000, p.504), and with C
R
1
2
3
C
0
(Pope, 2000,
Eq. (12.62)), which results from relating Rottas model with the
SLM, we obtain for the Rotta model constant C
R
= 4.15. The adopted
RANS model coefcients are summarized in Table 1.
For the discretization of the Reynolds stress and e transport
equations, a grid starting at x
1
= 0 with 100 equally-spaced grid
cells of length 6 mm was applied. At the inow, boundary condi-
tions for the mean velocities (Fig. 1), the mass loading, the dissipa-
tion rate and the isotropic Reynolds stresses were applied.
2
For the
numerical solution of the discretized equations, the solver twoPha-
seEulerFoam, being part of the OpenFOAM 1.7.1 (2012) software li-
brary was used as a basis with the ke turbulence model replaced
by the outlined Reynolds-stress turbulence model and with Eq. (4)
included.
5. Results
In a rst step, the Reynolds-stress model implementation was
tested by simulating the turbulent water ow in the absence of
particles. The resulting decay in downstream direction x
1
of the
turbulent kinetic energy k and the turbulence frequency x e/k
is plotted in Fig. 2. More precisely, the inverse of k is plotted, which
leads to a close-to-linear evolution as a function of x
1
. Both distri-
butions are in good agreement with the experimental data of Poel-
ma et al. (2007, Table 2).
In a next step, the two-phase cases considered by Poelma et al.
(2007) were simulated to investigate the accuracy of the new RANS
formulation. In Fig. 3, results of the rst (q
p
/q = 3.8, St = 0.07,
U
m
= 0.44) and the third experiment (q
p
/q = 2.45, St = 0.12,
U
m
= 0.65) of Poelma et al. (2007) with the second parameter pair,
i.e., C
R
= 4.15 with C
a
= 28, are reported. The particle phase leads to
an enhancement of the Reynolds stress in streamwise direction,
i.e., u
0
1
u
0
1

, and a reduction in spanwise directions, i.e.,
u
0
2
u
0
2
i hu
0
3
u
0
3

. This behavior is accurately captured by the model
and there is good agreement with the experimental data. Similar
agreement was found as well for the rst parameter pair and the
other experiments 2, 4 and 5 with both parameter pairs.
6. Conclusions
In different numerical and experimental studies it was found
that particles or droplets that move with a mean velocity relative
to the uid phase induce a Reynolds-stress anisotropy. A new com-
pact model formulation that accounts for this effect in Reynolds-
stress-based RANS simulations was proposed and successfully val-
idated based on the experimental data by Poelma et al. (2007).
References
Balachandar, S., Eaton, J.K., 2010. Turbulent dispersed multiphase ow. Annu. Rev.
Fluid Mech. 42, 111133.
Poelma, C., Westerweel, J., Ooms, G., 2007. Particle-uid interactions in grid-
generated turbulence. J. Fluid Mech. 589, 315351.
Geiss, S., Dreizler, A., Stojanovic, Z., Chrigui, M., Sadiki, A., Janicka, J., 2004.
Investigation of turbulence modication in a non-reactive two-phase ow. Exp.
Fluids 36, 344354.
Elghobashi, S., Truesdell, G.C., 1993. On the two-way interaction between
homogeneous turbulence and dispersed solid particles. i: Turbulence
modication. Phys. Fluids A 5, 1790.
Meyer, D.W., 2012. Modelling of turbulence modulation in particle- or droplet-
laden ows. J. Fluid Mech. 706, 251273.
Bear, J., 1972. Dynamics of Fluids in Porous Media. American Elsevier Pub. Co., New
York.
Pope, S.B., 2000. Turbulent Flows. Cambridge University Press, Cambridge.
The OpenFOAM Foundation, Open source eld operation and manipulation
(openfoam) c++ libraries, 2012.
2
Values are provided in (Meyer, 2012, Table 2) with the turbulence frequency
x e/k.
(a)
(b)
Fig. 3. Validation of Reynolds-stress spatial development resulting from two-phase
ow simulations (lines) and experiments 1 (panel a) and 3 (panel b) by Poelma et al.
(2007, Figs. 18 and 16) (symbols). hu1i
2
=hu
0
1
u
0
1
i (dash-dotted red line and red
squares), hu1i
2
=hu
0
3
u
0
3
i (dashed blue line and blue dots), and hu
1
i
2
/(2k/3) (solid black
line). (For interpretation of the references to color in this gure legend, the reader is
referred to the web version of this article.)
Y. Reinhardt et al. / International Journal of Multiphase Flow 56 (2013) 13 3

You might also like